Nature Communications: diffusion models emerge from dense associative memory, with three computational phases
DimaKrotov · x · 2026-10-29
A Nature Communications study by an IBM/RPI team including Dmitry Krotov analyzes diffusion models through the Dense Associative Memory lens (generalized Hopfield networks). Memories are stored at local minima of the energy landscape; when training data exceeds critical storage capacity, new local minima — spurious states — emerge. The authors identify three computational phases: memorization, generalization, and spurious pattern generation, characterizing all three via basins of attraction and curvature measurements, with the analysis extended to Stable Diffusion.
More from Research
- 3,957-task study finds no reliable way to combine human and AI oversight yet — sethlazar · 2026-10-29
- New preprint measures 'articulation gaps': how much knowledge in humans and AI is tacit — sanmikoyejo · 2026-10-29
- Peer Review Is the Least Articulable Field: Five Findings on Tacit Knowledge — sanmikoyejo · 2026-10-29
- New Prompt Optimization Beats GEPA by ~14% and Estimates Performance Bounds — sanmikoyejo · 2026-10-29
- Borrowing Capture-Recapture from Ecology to Bound Rubric Performance — sanmikoyejo · 2026-10-29
- VAT: Measuring Tacit Knowledge with Verifiable, Articulable, and Dense Models — sanmikoyejo · 2026-10-29